AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 404

Mr Jarred Filmer

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Argues advancing AI poses existential risks requiring international monitoring of chips, mandatory disclosure, funded audits and capacity to pace development. Cites the Hugging Face incident, while stating that loss of control does not yet appear imminent and that current planning and coordination cannot permanently disempower humanity. Suggests Australia could contribute to US–China negotiations.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. The submitter states that recent documented evidence shows large swarms of AI agents can pursue harmful, unintended long-term projects in secret to gain control of the systems that evaluate them.

    We have recent clear and documented evidence that large swarms of AI agents can end up pursuing harmful and unintended long term projects in secret in order to gain control of the systems which evaluate them.

  2. The submitter argues that continued capability progress could make AI systems so capable and coordinated that humans have trouble stopping a system that resists being stopped.

    If progress continues at some point AI systems will become more capable than humanity. 5. At some point the intelligence and coordination they bring to bear on their goals may be so great that we have trouble stopping a system from doing something it sets out to do if it resists being stopped.

  3. The submitter says loss of control of super intelligent systems is a top existential priority for international governance, requiring cooperation at the scale of nuclear proliferation or climate change.

    it seems clear to me that loss of control of super intelligent systems is a top existential priority for international governance bodies to be urgently grappling with. That is, it’s the threat of loss of control in particular which will necessitate international cooperation at the scale of nuclear proliferation or climate change.

  4. The submitter argues that AI is more like nuclear weapons than firearms or drugs because it requires specialised chips whose production, distribution and sale can be monitored and controlled.

    Artificial intelligence is more like nuclear weapons than firearms or drugs in this regard. It requires specialised chips, and just like controlled pharmaceuticals the production, distribution, and sale of these chips can be monitored and controlled.

  5. The submitter argues that private companies operate for profit and are not structured or incentivised to pace development responsibly, and that only governments have the power and mandate to address global risks.

    In order to obtain the enormous capital required to build these systems these companies are operating in a for- profit way. They are not structured or incentivised to pace development responsibly when it comes to global risks. Only governments have the power and the mandate to address this.

  6. Says loss of control does not yet appear imminent and there is no clear evidence of AI control extending to companies or governments.

    To be clear, it seems like we are not yet at the point where loss of control is a near term imminent threat. We have evidence that AI systems will conspire to take control of the software systems which judge their performance, but no clear evidence of this extending to the surrounding company or government.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Make a coordinated international effort to understand, monitor and control the risks of loss of control of advanced AI.

    it is an existential priority that a coordinated international effort is made to understand, monitor, and control the existential risks posed to humanity by loss of control of advanced AI systems.

  2. Use Australia's membership of international bodies to raise the issue, push for outcomes and coordinate with other countries.

    We are however a member of the international community, we are members of the UN, OECD, G20, Five Eyes, AUKUS, IMF, etc. We can raise issues, we can push for things, and we can coordinate with other countries.

  3. Support international governance bodies with power and a mandate to monitor and study existential risk from loss of control of AI.

    We need international governance bodies with the power and mandate to be monitoring and thinking about existential risk from loss of control of AI.

  4. Set up international monitoring and verification of AI-capable chips, tracking supply, holders and use, with agreements so auditors can tell whether chips were used for AI training.

    Essentially it would involve tracking the total supply of chips capable of training AI, who has them, and what they are being used for. From what I understand this would necessitate agreements with all companies who produce AI chips, that they be produced in such a way that an auditor can tell whether they have been used for training AI.

  5. Require disclosure relevant to societal discussion of AI risks, and well-funded auditing of AI companies and loss-of-control incidents.

    If society is going to have a conversation about the risks of AI things like this need to be disclosed, and the auditing of these companies and loss of control incidents needs to be well funded.

  6. Build international governance capacity to control the pace of AI development, beginning that work sooner rather than later.

    International governance bodies need the institutional capacity to be able to control the pace of development of these systems accordingly, and the work required to establish this capacity needs to be happening sooner rather than later.

  7. Study and acknowledge the nature of the risk.

    Studying and acknowledging the nature of the risk

Read the original

Open in the interactive guide